Comments on the definition of the Q2 parameter for QSAR validation

Area of Science:

• Quantitative Structure-Activity Relationship (QSAR) modeling
• Cheminformatics
• Computational chemistry

Background:

• Evaluating the predictive ability of QSAR models is crucial for reliable drug discovery and chemical safety assessments.
• Existing methods for external validation, such as those in OECD guidelines, use specific formulas for the predictive squared correlation coefficient (Q2).
• These formulas often rely on the sum of squares (SS) of the external test set relative to either the training set mean or the test set mean.

Discussion:

• This paper critically examines established QSAR validation metrics, specifically the predictive squared correlation coefficient (Q2).
• It analyzes two common Q2 calculation methods: one referencing the training set mean and another referencing the test set mean.
• A novel Q2 formula is proposed and evaluated, utilizing SS relative to the training set mean over the training set itself.

Key Insights:

• The choice of reference mean (training set vs. test set) significantly impacts Q2 values and the assessment of predictive ability.
• The proposed Q2 formula offers an alternative perspective by considering deviations within the training set, potentially providing a more robust measure of internal consistency.
• Understanding these variations is essential for accurate QSAR model performance evaluation.

Outlook:

• Further research is needed to validate the proposed Q2 formula across diverse QSAR datasets and model types.
• Comparative studies will clarify the advantages and limitations of each Q2 calculation method in different cheminformatics contexts.
• This work contributes to developing standardized and more reliable QSAR validation protocols for regulatory applications.